Soft sensor for dynamic fluid level of beam pump unit based on multiple LS-SVM models

It is difficult to realize continuous measurement of down-hole dynamic fluid level of beam pump unit in practical oilfield production. To solve this problem, a soft senor method based on multiple LS-SVM (Least Squares Support Vector Machine) models is proposed in this paper. Many directly measurable parameters of the pumping system are selected as instrumental variables. Improved ISODATA (Iterative Self-Organizing Data Analysis Technique) dynamic clustering algorithm is used to divide the training samples into different groups which are used to train different LS-SVM sub models. The weights of different sub models are calculated according to the mean square error of different groups of training samples. The testing samples are calculated by different sub models to obtain different outputs, and the final result is got by the weighted average calculation according to their weights. The radial basis function is used as the kernel function. PSO (Particle Swarm Optimization) algorithm is used to choose the best regularization parameter C and Gaussian kernel parameter σ of different LS-SVM sub models. Case study shows that the proposed method in this paper has better effectiveness to realize soft sensor for down-hole dynamic fluid level of beam pump unit.